The Mystery LandThe Mystery Land 

This first shot was my introduction to the advanced techniques of the course. We started by applying the chroma key, working on the character's edges and solving small imperfections in the plate. Here I understood the critical importance of chroma subsampling to extract a clean and professional key. Then we moved on to integrate the background. I learned a very useful technique to match the BG perspective with the FG, taking into account that both images had been captured with different cameras and lenses. The formula used was: (FG Focal length) / (BG Focal length × (FG sensor / BG sensor)) This determines the exact scale at which the BG needs to be enlarged to match in perspective with the actor, something especially important in a close-up like this.

Before

After

In this shot the challenges increased. The background needed much more work: a ground area had to be added and the entire environment was part of a complex DMP. As in the previous shot, I had to adjust the perspective, but I also projected the ground on a horizontal plane and the background on a vertical plane to simulate real depth. This allowed me to generate a depth channel that I used to apply realistic z-defocus, fundamental for a believable integration. On the character, the chroma key was more difficult: the actor came out in some parts of the green background and additional rotoscoping was necessary. Additionally, we had to create a 3D layout inside Nuke using the real lidar from the location. This was key to validate the camera's 3D tracking, which in the original plate had a lot of shake. Comparing the 3D layout with the real material allowed adjusting and correcting the track with professional precision.

This was, without a doubt, the most complex shot of the project. Here I had to apply absolutely EVERYTHING I learned: • 3D layout to verify the camera • 3D tracking • Advanced rotoscoping where chroma was no longer sufficient • Edge correction • Fine integration of the actor To that work was added a complete CG environment that needed coherent lookdev to integrate it correctly with a huge DMP, composed of many layers and elements. This DMP was heavy and complex both artistically and technically, and required a brutal effort of control, order, and consistency.

This shot is one of my favorites from the project. Here I learned to perform truly professional 3D tracking, orienting, scaling, and positioning the 3D camera with real measurements thanks to the location's lidar and various tracking support techniques. The result was an extremely precise camera, without slippage and perfectly aligned to the environment. Thanks to this tracking, I was able to perform a terrain clean-up: remove a mound and modify several dirt paths, since two CG ships were going to be integrated on top of them. I did the lookdev of the ships so they would integrate with the real plate and added additional 2D elements to reinforce their physical presence. The crashed ship had earth around it that had to be carefully integrated with that of the plate. Through various nodes that broke masks and generated irregularities, I achieved a believable interaction between the CG and the real environment. I also performed a sky replacement, taking advantage of the 3D camera and some keyers. The large foreground statue had a very CG look and it was essential that it appear captured in camera, so I polished it meticulously using Kernel nodes and adjusting it perfectly through the 3D camera. Finally, I performed black matching on all shots, an essential step for final integration and general color grading.